Video Encoding
Our video encoding testing comprises of comparing CPU only Video Encoding versus GPU accelerated Video Encoding over several tests.
AMD, INTEL and NVIDIA all now offer full hardware video encode acceleration which depending on the implementation is either optimised for real-time encoding or highly accelerated encoding.
Our Movie Maker 2012 test comprises a simple project were we import Big Buck Bunny 1080p sample clip, add some titles, transitions, trim the video and export the finished result. Movie Maker 2012 utilises DirectX and through this the GPU's video engine. AMD's engine, called VCE is optimised to be used in conjunction with OpenCL or for real time encoding and is almost twice slower than AMD or NVIDIA, which is unacceptable for a full bitstream h264 encoder in hardware.
Moving on to Cyberlink MediaEspresso, comparing GPUs suing an A-B test, although there is inconsistent with how MediaEspresso uses GPU acceleration, AMD is constantly the slowest here and is significantly slower than CPU only encoding. Again unacceptable.
During its 2013 developers conference, AMD described how it has applied two of its engineers to rewrite both the popular Handbrake and x264 video encoders in an effort to not only optimise for their products but provide significant speedup via OpenCL.
Only specific parts of the encoding pipeline such as scaling are accelerated using OpenCL in Handbrake, however we do see a significant speed up of 40 frames per second over CPU only encoding with Radeon. So Handbrake and OpenCL finally work. Intel QuickSync has always been ridiculously fast in handbrake thanks to a custom codec suited for the hardware unit.
The Normal preset does not scale the image so we see no speed up.
